This is an official implemention for “SiamCorners: Siamese Corner Networks for VisualTracking”. The code are available here now.

The overview of our SiamCorners architecture, which includes the Siamese feature extractor followed by the top-left corner and bottom-right corner branches in parallel.

• Python 3.7
• PyTorch 1.0.0
• numpy
• CUDA 10
• skimage
• matplotlib
• GCC 4.9.2 or above

Prepare training dataset, detailed preparations are listed in training_dataset directory.

• VID
• YOUTUBEBB (BaiduYun, extract code: h964.)
• DET
• COCO
• LaSOT
• GOT10k

Compile the C++ implementation of the corner pooling layers. (GCC4.9.2 or above is required.)

cd <SiamCorners dir>/pysot/models/corners/py_utils/_cpools
python setup.py install --user

Compile the NMS code which are originally from Faster R-CNN and Soft-NMS.

cd <SiamCorners dir>/pysot/tracker/external
make

CUDA_VISIBLE_DEVICES=0,1
python -m torch.distributed.launch \
    --nproc_per_node=2 \
    --master_port=2333 \
    ../../tools/train.py --cfg config.yaml

python ../tools/test.py 

If you're using this code in a publication, please cite our paper.

@InProceedings{SiamCorners,
author = {Kai Yang, Zhenyu He, Wenjie Pei, Zikun Zhou, Xin Li, Di Yuan and Haijun Zhang},
title = {SiamCorners: Siamese Corner Networks for Visual Tracking},
booktitle = {IEEE Transactions on Multimedia},
month = {April},
year = {2021}
}

Our anchor-free tracker is based on PySot and CornerNet. We sincerely thank the authors Bo Li and Hei Law for providing these great works.

If you have any questions, please feel free to contact [email protected]